QVAC SDK has released version 0.11.0, which now supports the Qwen 3.5 and Qwen 3.6 models, as well as Gemma 4, significantly enhancing local computing capabilities and visual workflows. This update allows users to leverage multi-GPU support for running larger models locally, and introduces features like multi-image conditioning for improved style mixing and on-device upscaling for high-quality images. These advancements align with a growing trend in AI tooling that prioritizes on-device processing to enhance privacy and reduce reliance on cloud services, catering to the increasing demand for versatile multimodal creative workflows.

QVAC: QVAC is Tether’s AI research and software initiative focused on local-first, on-device computing and developer tools for running and building AI workflows. In this update, QVAC SDK 0.11.0 adds expanded support for newer model families, multi-GPU execution, and visual generation features that are designed to run on users’ own hardware.
Qwen: Qwen is Alibaba’s family of open-weight language models used widely for chat, coding, and multimodal applications. It is relevant here because the latest QVAC SDK release adds support for Qwen 3.5 and Qwen 3.6, expanding the set of models developers can run locally through the framework.
Gemma: Gemma is Google’s family of open models aimed at efficient deployment across developer and edge environments. In this news, QVAC SDK adds support for Gemma 4, indicating the framework is broadening compatibility with newer open model releases for local inference.

Local AI: Recent AI tooling has increasingly emphasized running models on-device to improve privacy, reduce reliance on cloud services, and enable offline use.
Visual Workflows: Modern AI SDKs are expanding beyond text generation to include image conditioning and upscaling, reflecting growing demand for multimodal creative workflows.
Model Compatibility: Framework updates that add support for newer open-weight model families often aim to make local AI stacks more practical for builders who want flexibility across different model ecosystems.